发明名称 VOICE RECOGNITION METHOD, DEVICE THEREFOR, AND COMPUTER CONTROL DEVICE
摘要 PROBLEM TO BE SOLVED: To reduce the searching space for unknown word or the memory quantity for unknown word in a voice recognition processing capable of detecting an unknown word. SOLUTION: The state transition model for unknown word set by the number of states and each transition probability between the states is described in a HMM 104. A second output probability calculating part 105 gains the state where a voice data has the highest likelihood every time of a plurality of states used in the state transition model for known word used for voice recognition of known word. The result is applied to the state transition model for unknown word described in the HMM 104 to gain the state transition model for unknown word. A first output probability calculating part 103 determines the likelihood of the state transition model for known word. A language searching part 107 executes a language searching processing for the part where the presence of unknown word is allowed in a grammer.dictionary 106 by use of the likelihood from the first and second output probability calculating parts.
申请公布号 JPH09127978(A) 申请公布日期 1997.05.16
申请号 JP19950285344 申请日期 1995.11.01
申请人 CANON INC 发明人 KOMORI YASUHIRO;YAMADA MASAAKI;OHORA YASUNORI
分类号 G10L15/10;G10L15/06;G10L15/08;G10L15/14;G10L15/18;G10L15/20;G10L15/28;(IPC1-7):G10L3/00;G10L3/00 主分类号 G10L15/10
代理机构 代理人
主权项
地址